Anthropic: latest AI news
All Oossa news about Anthropic, newest first.
Known for: Claude, Claude Code
Follow via RSSToday
Yesterday
Tools & agents
Vercel releases durable-harness to keep coding agents running after crashes
Vercel open‑sources durable-harness, a package that logs model and tool calls so sessions with Pi, OpenCode, Claude Code and Codex can resume after crashes or redeploys.
Tools & agents
Anthropic launches Claude Duet AI agent for professionals
Anthropic released Claude Duet, a research‑preview AI assistant for small businesses, available on iOS, Android and the web.
Tools & agents
Anrok launches MCP to link tax data with AI assistants
Anrok's new MCP server lets finance teams access and update tax information directly from tools like ChatGPT and Claude.
Tools & agents
Anthropic updates API and SDKs to version 2026-10-21
New API version released with updated client libraries and a license change for the SDKs and CLI.
Models
Unstructured launches Transform models that out‑perform GPT‑6, Claude and Gemini
In tests on 224 business pages, Unstructured's new Transform models made far fewer table errors and hallucinations than leading AI models.
Policy & law
Meta leads big‑tech lobbying on AI antitrust rules
Meta, OpenAI and Anthropic filed lobbying reports showing they pushed Washington on AI information‑sharing rules in Q3 2026.
Business
OpenEvidence AI startup valued at $15 billion sees mixed clinical results
The ad‑funded AI company behind OpenEvidence claims strong evidence‑based recommendations, but independent studies show it trails general‑purpose models.
Safety
Hackers use older AI models to bypass safeguards in attacks
CrowdStrike reports four groups exploiting open AI tools like Anthropic Claude and Deepseek to run autonomous hacks between May and August 2026.
Safety
Zero Networks Tests AI Attack in Anthropic Verification Program
Zero Networks reports that a frontier AI model could not breach its microsegmented network after gaining admin access.
Policy & law
US bans Deepseek AI products; China says it’s discrimination
The US Department of Homeland Security ordered federal agencies to drop Deepseek tools, while China blocks ChatGPT, Claude and Gemini at home.
Business
Anthropic plans $750 billion compute spend
Anthropic’s prospectus shows a $750 billion investment in AI hardware as it prepares for a possible $2 trillion IPO.
Models
Anthropic’s Claude designs protein binders against deadly scorpion toxins
Claude created over 300,000 protein designs in 24 hours, with 17 showing binding activity in lab tests.
Tuesday, October 20
Models
Inception launches Mercury 3, a diffusion model up to 8× faster
Inception AI’s new Mercury 3 model claims to generate text four to eight times faster than OpenAI’s GPT‑6 Luna and Anthropic’s Claude Haiku 5.5.
Safety
Anthropic drafts AI safety bill that could block models posing deadly risk
Anthropic proposes a Congress bill that would halt AI model releases if they could cause 50 deaths or $1 billion in damage, using independent evaluators.
Business
Eric Schmidt owns about 1% of Anthropic, half of CEO's stake
Bloomberg says Schmidt's investment gives him a 1% share in Anthropic, compared with CEO Dario Amodei's 2% stake.
Models
Anthropic rolls out stronger anti‑distillation safeguards
Anthropic says its new defenses cut extracted reasoning data by 10‑100×, though some legit users must update integrations and may be charged for blocked requests.
Models
Anthropic’s Claude writes the first AI‑planned Mars rover drive
On Dec 8 and Dec 10 2025 NASA used Claude to plot a 400‑m route for Perseverance, cutting planning time roughly in half.
Models
llama.cpp update adds id_slot handling for Anthropic API
The Oct 20 2026 release lets servers pin sessions to specific KV cache slots, avoiding full re‑prefill on Anthropic /v1/messages calls.
Policy & law
Anthropic’s draft AI safety bill draws criticism
A leaked proposal Anthropic has pitched to lawmakers faces criticism over safety requirements and limits on state laws. The company says it plans changes.
Business
Anthropic and Nscale in talks for another $45 billion compute partnership
Bloomberg says the two AI firms may soon announce a second $45 billion deal as they prepare for IPOs.
Monday, October 19
Models
Amazon Bedrock adds JSON‑schema outputs for Claude models
All Anthropic Claude models on Bedrock can now return schema‑validated JSON, cutting parsing errors and retry logic.
Business
Anthropic’s IPO faces uncertainty from a private trust
A little‑known trust could affect the timing and terms of Anthropic’s upcoming public offering.
Models
Anthropic proposes equity fund for AI‑displaced workers
Anthropic says a new LLC subsidiary will allocate about 20% of its equity to a fund aimed at helping workers affected by AI, while Elon Musk touts giving SpaceX shares to children.
Safety
Claude Code flags XMRig miner on developer’s Mac
A developer says Claude Code spotted XMRig using about 1,175% CPU on an overheating Mac. The account does not explain how the miner got there.
Tools & agents
MIT researchers test Claude in a virtual island experiment
Claude Opus 4.7 agents explored a simulated island and invented games. Researchers say the activity helped them do later tasks, but whether it counts as play is disputed.
Sunday, October 18
Business
Anthropic pledges 5% of its shares to a public benefit fund
The AI startup will lock up a portion of its equity to fund nonprofit work and study AI’s impact on jobs.
Business
Anthropic lets watchdog METR use its tools like staff
Anthropic has given six METR researchers company hardware, office space and internal‑tool access, matching what its own risk team sees.
Tools & agents
Anthropic cuts junior hires as AI agents work six times more hours
Anthropic’s internal study shows AI agents logged six times the work hours of humans by August 2026, while junior technical roles fell from 21% to 14% of staff.
Safety
Anthropic pushes AI safety bill that triggers only on massive harm
Anthropic is lobbying a bill that would let the Commerce Department block AI releases only if they could cause at least 50 deaths or $1 billion in damage.
